Rerun

Rerun is a specialized open-source SDK and visualizer purpose-built for logging and exploring computer vision and robotics data over time. Unlike general-purpose logging or dashboarding tools, Rerun is designed from the ground up to handle the multimodal, spatial, and temporal complexity that defines modern embodied AI systems. Its core value proposition is straightforward: with minimal code, engineers and researchers can capture internal state—camera frames, 3D point clouds, transforms, scalars, text logs—and then interactively scrub through that data in a dedicated viewer. This makes it a practical debugging and validation tool for anyone building perception pipelines, sensor fusion stacks, or spatial AI models.

Where Rerun stands out is in its combination of logging ergonomics and visualization fidelity. The SDK, available in C++, Python, and Rust, lets users log heterogeneous data types to a single timeline with just a few lines of code. The visualizer then renders that data with temporal controls, allowing frame-by-frame inspection, side-by-side comparisons, and playback at variable speeds. This tight loop between logging and viewing is critical for debugging nonlinear failures in robotics or vision systems—where a bug might manifest only under specific sensor conditions or after long runtimes. Rerun’s Rust-based architecture ensures the viewer remains responsive even when handling large logs, a practical advantage over slower alternatives.

The tool fits best into workflows that already involve iterative development of spatial AI systems. For a computer vision engineer debugging a 3D reconstruction pipeline, Rerun can log intermediate depth maps, camera poses, and reconstructed meshes, then let the engineer scrub through the sequence to pinpoint where geometry drifts. For a robotics engineer validating sensor fusion, Rerun can ingest LiDAR scans, IMU data, and odometry estimates, overlaying them in a shared 3D view to check alignment. Researchers preparing paper walkthroughs can log experimental data and export interactive visualizations that convey temporal dynamics more effectively than static figures. The Aria Dataset Explorer integration shows how Rerun can serve as a front-end for large public datasets, but its primary strength is in ad-hoc, custom logging.

Who benefits most? Teams building custom perception or robotics stacks—especially those dealing with multi-sensor setups or state estimation—will find Rerun immediately useful. It reduces the friction of setting up visual debugging, which often requires stitching together separate tools for different data types. Solo researchers and small teams can adopt it quickly due to its open-source nature and straightforward API. Larger organizations may leverage the managed infrastructure tier for centralized ingestion and streaming, though this introduces operational overhead and cost considerations not present in the self-hosted version.

Limitations matter. Rerun is laser-focused on spatial and embodied AI domains; it is not a general-purpose log aggregator or a model training monitor. It lacks built-in analytics, alerting, or dashboarding for business metrics. The visualizer is a thick client, not a web dashboard, which limits remote collaboration unless combined with streaming infrastructure. Furthermore, Rerun requires integration into an existing codebase—it is a logging library, not a plug-in. Teams must add logging calls to their code, which can be a barrier for legacy systems or closed-source components. There is also no native support for real-time streaming from embedded devices; data is typically logged to files and viewed post-hoc.

For a practical buyer or operator, the decision to adopt Rerun hinges on whether your debugging pain is specifically about understanding temporal, multimodal spatial data. If you are struggling to correlate camera images with LiDAR points and robot poses across a time series, Rerun offers a purpose-built solution that likely outperforms cobbling together separate viewers. If your needs are broader—covering system logs, application performance, or business KPIs—a more general observability platform may be appropriate. Rerun is best evaluated by running its quickstart guide with your own data; the value becomes immediately apparent when you see your first logged sequence play back in the viewer. The open-source core ensures low risk for experimentation, and the community Discord provides responsive support for integration issues. Ultimately, Rerun fills a genuine gap in the developer toolchain for spatial AI, and its focus and performance make it a serious contender for teams willing to invest a few hours in integration.